A Deep Introduction to AI Based Software Defect Prediction (SDP) and its Current Challenges

Abstract

Software systems are driving the quality of human life. Annual investment in software systems development is exceeding 4 trillion USD. Given the increasing complexity of software products and cognitive nature of software development process, the software defects are on the rise resulting in poor quality software that needs about 2.8 trillion USD to fix. To help improve software quality, the field of Software Defect Prediction (SDP) emerged. Its novel attempt to isolate defective software units enables defect removal as well as better utilization of resources in software development and maintenance activities. Having begun with a humble set of statistical models, SDP now employs sophisticated machine learning techniques delivering good results. SDP is now considering use of highly accurate search-based techniques for defects prediction. Also SDP is emerging as a formal methodology having its own process, model and evaluation criteria. Still, the SDP is troubled by a variety of problems such as diversity of datasets, difficulty in selecting ideal set of features to predict software defects, lack of robust methodology to build SDP models and non-availability of feature rich, cross-project industry (commercial) datasets to build ideal SDP models for use across multiple projects etc. There is a plenty of scope to strengthen SDP and also take measures to promote its practical use within the industry by simplifying SDP models and quantifying their outputs and benefits.